What if one of the most important discoveries in modern space science was not made by a billion-dollar telescope, but hidden in plain sight, waiting for someone to look differently?

That is exactly what happened when a Pasadena high school student used artificial intelligence to unlock a cosmic treasure trove buried inside years of overlooked data. The result is staggering. About 1.5 million previously unrecognized objects in space have now been identified, forcing scientists to confront a powerful reality. The sky was never fully mapped.

Matteo Matthew Paz built an AI system that reanalyzed infrared data from NASA’s NEOWISE mission and revealed patterns no human had been able to systematically detect. His peer reviewed study, published in December 2024 in The Astronomical Journal, is now being seen as a turning point in how scientists approach massive datasets.

A Summer Project That Turned Into a National Breakthrough

What began as a summer experiment quickly evolved into something far bigger.

Paz started his work in 2022 through Caltech’s Planet Finder Academy, later continuing in a six week research program that paired students with professional mentors. His guide was astronomer Davy Kirkpatrick from Caltech’s Infrared Processing and Analysis Center.

The project rapidly outgrew its original scope. It ultimately earned Paz the top prize in the Regeneron Science Talent Search, along with a 250000 dollar award.

Instead of analyzing the sky piece by piece, Paz trained his system to recognize patterns automatically. With each iteration and mentor feedback session, the model improved. He later described those meetings with a revealing touch of humor, saying each one was "10% work and 90% us just chatting."

A Space Mission Full of Hidden Signals

NEOWISE was designed to track asteroids and near Earth objects. But while scanning for threats, it was quietly collecting something even more valuable. Time based infrared signals from distant objects that flicker, pulse, brighten, or dim.

These include quasars, exploding stars, and eclipsing binaries. In other words, a dynamic universe hidden inside static data.

The problem was never lack of information. It was scale.

The NEOWISE database contains nearly 200 billion individual detections collected over more than a decade. The full dataset reaches about 200 terabytes. That volume makes manual analysis nearly impossible.

Kirkpatrick captured the challenge in a way many can relate to. The team was "creeping up towards 200 billion rows", turning even small tests into slow, painstaking efforts.

How VARnet Spotted What Humans Missed

Paz’s solution was a system called VARnet, designed to detect subtle changes over time in astronomical data.

It works by analyzing light curves, which track how an object’s brightness changes. The system breaks those signals into patterns across different time scales, separating real variability from random noise.

The approach combines wavelet decomposition with Fourier based feature extraction, then feeds the results into a neural network trained to recognize meaningful signals.

The performance is striking. Processing takes less than 53 microseconds per source on a modern graphics processor, making it scalable across massive datasets.

Accuracy also holds up under pressure. VARnet achieved an F1 score of 0.91 in tests involving real infrared variables, showing it can reliably classify different types of cosmic changes.

From Raw Data to a Usable Cosmic Map

The discovery of 1.5 million new objects does not mean they suddenly appeared. It means their variability had never been properly identified or categorized at scale.

Paz’s system created a catalog of about 1.9 million infrared variable objects. Roughly 1.5 million of them are newly recognized in this structured way.

The catalog also organizes these objects into 10 distinct categories, allowing scientists to quickly focus on specific phenomena.

There is a major observational advantage as well. Infrared data can penetrate dust clouds that block visible light, meaning the system can detect activity in regions where traditional surveys struggle. Some of these signals were effectively invisible until now.

Why This Discovery Matters Beyond Space

At first glance, mapping distant stars may seem far removed from everyday life. But the underlying method has powerful real world applications.

The key is time series analysis. The same technique used to track flickering stars can be applied to Earth based data, from pollution cycles to climate patterns.

Paz himself pointed to this connection, explaining that his model could "study atmospheric effects such as pollution" by analyzing repeating patterns shaped by seasons and daily cycles.

This opens the door to better environmental monitoring, earlier warnings, and more precise tracking of subtle changes that impact human health and safety.

The Hidden Cost of a Data Driven Future

There is another side to this breakthrough that cannot be ignored.

As artificial intelligence becomes central to scientific discovery, the energy required to power these systems is becoming part of the equation. Massive computations come with real environmental costs, from electricity usage to infrastructure demands.

The same technology that helps monitor the planet also contributes to its energy footprint.

That tension is now part of the broader conversation.

A New Era of Discovery Has Begun

The most important takeaway is not just the number of objects discovered. It is the shift in how discovery happens.

Vast archives of data are no longer static records. They are dynamic resources waiting for the right tools to unlock them.

What Paz demonstrated is both simple and profound. The next major breakthrough may not require a new telescope. It may already exist, hidden in data we thought we understood.

All it takes is a new way of seeing.